On Thu, 2002-09-05 at 11:39, Mark Stewart wrote:
> Hi Lyvim,
> 

> > Synopsis:  Slaving the Cdrom should be OK.
> >
> 
> That makes sense as I recall civileme describing a quite elaborate soft RAID
> setup on a board with only 2 IDE channels.
> 
> But am I remembering correctly that having a slower device (my CDROM is
> ATA33 while the disk is at least ATA100) on the same channel mandates that
> both drives operate at the lower transfer rate? SCSI, at least once upon a
> time,worked like this. Has this been an issue for you (balancing the
> increased speed of the raid against the slower disk speed)? Or this just the
> case when both devices are being accessed at the same time?
 
I am fairly certain that caveats you refer to for SCSI don't apply to
IDE.  For one thing, I've never noticed a slowdown myself with
dissimilar devices on the IDE bus.  For another, devices are recognized
individually (type and speed) when the system boots up.  In addition, if
there was a problem, then anybody with an ATA100 drive would be wasting
their money if they bought a CDROM and put it on the same bus.  I
believe that one of the positive things about IDE (besides being cheap)
is that the devices are dealt with pretty much on an individual basis.

The problem comes in when you have a raid array in which one of the
devices is a slave.  When that happens, the master device must oversee
data transactions (during stripe operations) not only for itself, but
also for it's slave.  This makes the concept of dividing the work
between two peripheral devices impotent.  When you *do* have only master
devices in the array, it equates to a (x * 100)% performance increase
(if it's Raid 0) where x is the total number of devices.  Because the
work that was done by one is now done by several.

The Highpoint and generic IDE buses can be maxed out throughputwise, but
not, I think, by only one master device per channel.

 
> 
> Right and I'm only bothering with this because I need this box to be as fast
> i/o wise as it can be and the hardware budget has already been spent for the
> year...

I know where you're coming from..been there and got the t-shirt and sweatband.  You 
are headed in the right direction.

The cool thing is that your generic IDE channels can be freed up for
experimenting with other peripherals; and you won't ever have to muck
around with the channels your primary devices are on.


> This is what I figured I ought to be doing--leaving the RAID disabled in
> BIOS--but, as I indicated, DiskDrake does not see the Highpoint controller's
> IDE channels at all. Do you know if  there a secret handshake I have to give
> DiskDrake to get it to see the controller?

This is worrisome.  I can't offer advice on the 372, since I don't have
one; but I'm surprised that Diskdrake is croaking.  Have you tried an
install with the drives already partitioned, and highpoint raid
deactivated?
 
> cheers,
> ::mark
> 
> P.S. I dug around and found the Configure.help file in
> /usr/src/Linux/Documentation for 8.2's default kernel (2.4.8) and it talks a
> bit about the Highpoint controllers pre-372. I'm beginning to think that the
> hpt372 just isn't supported yet. At least not by name. Highpoint has a page
> talking about the Linux version of their hardware RAID driver which suggests
> the driver for the 372 and the 370 are mostly the same. But I'm not entirely
> convinced because with RAID enabled in BIOS the installer spews out some
> messages indicating it thinks it's dealing with a Highpoint 370 right before
> it locks up.
> 

L8r,

LX
 

-- 
���������������������������������������������������
Kernel  2.4.18-6mdk     Mandrake Linux  8.2
Enlightenment 0.16.5-11mdk    Evolution  1.0.2-5mdk
Registered Linux User #268899 http://counter.li.org/
���������������������������������������������������


Want to buy your Pack or Services from MandrakeSoft? 
Go to http://www.mandrakestore.com

Reply via email to